On the compensation for chance agreement in image classification accuracy assessment, Photogram

In the assessment of image classification accuracy with the kappa coefficient, the degree of chance agreement may be overestimated. Two kappa-like approaches which compensate more appropriately for the degree of chance agreement are discussed. Deriving these coefficients for a classification error matrix which had a percentage accuracy of 76.6 percent produced coefficients of approximately 0.69, higher than the kappa coefficient of 0.60. Because these alternative measures make a more appropriate compensation for chance agreement, they may be more suitable for the assessment of image classification accuracy than the kappa coefficient.